Unsupervised morphological classification of QRS complexes

Compares different strategies for nonsupervised classification of QRS complexes and reports results on the MIT/BIH arrhythmia database in respect to discriminatory power and computational demand. One approach is based on a hierarchical cluster analysis procedure with three different feature sets consisting of coefficients of orthogonal series expansions. The second method uses a two step correlation technique. For several reasons, our results suggest a preferability of the second method as long as a moderate signal quality can be guaranteed